Automatic Musical Instrument Identification for Tones of Unknown Frequency
by
Robert Moore
Victororia University of Technology
Coauthors: Pietro Cerone(Victoria University of Technology), Tom Peachey (Monash University)

Automatic musical instrument identitification is a relatively straight forward task when considering tones of the same pitch. A more difficult task is to discriminate between musical tones of differing and unknown pitch. The intent of this paper is to outline a method that enables the identification of a musical tone of unknown pitch(fundamental frequency) and discuss some of the computational difficulties.

Each sample tone is a sound wave encasing a large amount of information about the nature of that particular musical tone. Consequently, before classification of tones can take place, some form of pre processing which highlights the most discriminating features must be performed. The goal of the pre-processing is to develop a characteristic signature for each class of musical tone under consideration that will enable classification. In this paper some computational difficulties relating to the pre-processing stage will be examined. The implimentation of this stage involves the use of the discrete Fourier transform to obtain a series of frequency spectrums for a tone of unknown frequency. The main problem in achieving this outcome is that accurate representation of a frequency spectrum at a particular time is dependent on first determining the fundamental frequency of the tone.

Computational difficulties relating to the high dimensionality of the data passed to the classification stage will also be considered, in particular, where the number of dimensions (variables) associated with each observation is greater than the total number of observations.

Date received: July 29, 1999

Copyright © 1999 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cadk-70.